Generative Adversarial Networks-Based Video Anomaly Detection Approaches

Despite the many approaches introduced so far for anomaly detection in video, the automatic detection of abnormal crowd events in video data remains a challenging problem. In this chapter, we introduce the generative adversarial networks (GANs)-based anomaly detection approaches. Different from the previous state of the art, GAN detects anomalies in video data by exploiting temporal dependencies and discovering the underlying distribution of the baseline, without making any assumptions on its nature, offering a powerful approach for anomaly detection in complex, difficult-to-model video data. Therefore, the generative models represent a type of promising approaches for anomaly detection, especially in face of the increasing complexity and ever-growing number of objects to monitor nowadays in video scenes.